[U-Boot] [PATCH 3/7] 83xx/85xx/86xx: Add ECC support

[Peter Tyser]

> Ok, here are my results, this is on a 8349EMDS-derived board. My
> 8349EMDS eval board doesn't have ECC memory.
> 
> 1) It might be nice to have something to print the current injection
> registers. It is not a big deal, anyone using this should be an expert
> anyway.

Thanks for the feedback.  I can add a printing of the current injection
values when "ecc inject" is ran if others would like.

> 2) ecc inject off didn't seem to work, see the following capture:
> 
> => ecc info
> No ECC errors have occurred
> => ecc inject low 0x1
> => ecc info
> 
> WARNING: ECC error in DDR Controller 0
>         Addr:   0x0_0ff7ae40
>         Data:   0x0fffdf9c_0ff7aed1     ECC:    0x81
>         Expect: 0x0fffdf9c_0ff7aed0     ECC:    0x81
>         Net:    DATA0
>         Syndrome: 0x3b
>         Single-Bit errors: 0x1e
>         Attrib: 0x01002001
>         Detect: 0x80000004 (MME, SBE) 
> 
> => ecc inject off
> 
> # Ok, now error injection is off, I still expect some errors to be
> # present in the error registers
> 
> => ecc info
> 
> WARNING: ECC error in DDR Controller 0
>         Addr:   0x0_0ff7ae1c
>         Data:   0x0fffdf9c_0ff7d2a1     ECC:    0xe4
>         Expect: 0x0fffdf9c_0ff7d2a0     ECC:    0xe4
>         Net:    DATA0
>         Syndrome: 0x3b
>         Single-Bit errors: 0xd1
>         Attrib: 0x01003001
>         Detect: 0x80000004 (MME, SBE) 
> 
> # And there was the error. Now, I don't expect any more errors to
> # be present, after all, injection is disabled.
> #
> # But there is one! Why?

I believe what's happening is:
1. You turn error injection on
2. Every time you perform a DRAM write, the value written has an ECC
error
3. You write to DRAM lots of times, in lots of locations
4. You turn error injection off
5. There are still lots of ECC errors residing in DRAM that you discover
later when you read from "corrupted" memory locations

So in theory, unless you scrub your memory, you might uncover lots more
ECC errors later.

As an easily reproducible example try:
> ecc inject low 1; mw.l 0x100000 0xbeefba11 0x800000; ecc inject off
> ecc info
> ecc info
> md 0x100000
> ecc info
> ecc info
> md 0x200000
...

The majority of the above ecc errors could be cleared by running the
following command with ecc injection off:
mw.l 0x100000 0xbeefba11 0x800000


> => ecc info
> 
> WARNING: ECC error in DDR Controller 0
>         Addr:   0x0_0fff8a0c
>         Data:   0x0fff8a00_0fff8a01     ECC:    0xff
>         Expect: 0x0fff8a00_0fff8a00     ECC:    0xff
>         Net:    DATA0
>         Syndrome: 0x3b
>         Single-Bit errors: 0x04
>         Attrib: 0x01003001
>         Detect: 0x00000000
> => 
> 
> # Note that I keep seeing ecc errors until I run the command:
> # ecc inject low 0

Hmm...  "ecc inject off" should have the same effect as "ecc inject low
0".  Is there a chance some of the ECC errors still remaining in DRAM
are the culprit?

> # Why did it take two runs of ecc info to clear all of the errors?

This is probably the same issue as above - lots errors are injected and
there's no saying when exactly they'll turn up.

> Other than the above strangeness, everything is working great on my 83xx
> board. I think the new output is pretty nice. It serves my purposes
> equally well to the old code.

Thanks for trying the changes out,
Peter